comparison vp3.c @ 11133:cd2956d08cc1 libavcodec

Don't pre-calculate first_pixel 3.6% faster on Elephants_Dream_HD-q7-aq7.ogg on my penryn

11132:449c12b6c3a0

} Coeff;

//FIXME split things out into their own arrays
typedef struct Vp3Fragment {
    Coeff *next_coeff;
    /* address of first pixel taking into account which plane the fragment
     * lives on as well as the plane stride */
    int first_pixel;
    /* this is the macroblock that the fragment belongs to */
    uint16_t macroblock;
    uint8_t coding_method;
    int8_t motion_x;
    int8_t motion_y;

    Vp3Fragment *all_fragments;
    uint8_t *coeff_counts;
    Coeff *coeffs;
    Coeff *next_coeff;
    int fragment_start[3];

    ScanTable scantable;

    /* tables */
    uint16_t coded_dc_scale_factor[64];

    /* this is a list of indexes into the all_fragments array indicating
     * which of the fragments are coded */
    int *coded_fragment_list;
    int coded_fragment_list_index;
    int pixel_addresses_initialized;

    /* track which fragments have already been decoded; called 'fast'
     * because this data structure avoids having to iterate through every
     * fragment in coded_fragment_list; once a fragment has been fully
     * decoded, it is removed from this list */

    int height          = s->fragment_height >> !!plane;
    int fragment        = s->fragment_start        [plane] + ystart * width;
    int stride          = s->current_frame.linesize[plane];
    uint8_t *plane_data = s->current_frame.data    [plane];
    if (!s->flipped_image) stride = -stride;

    for (y = ystart; y < yend; y++) {

        for (x = 0; x < width; x++) {
            /* This code basically just deblocks on the edges of coded blocks.

            if( s->all_fragments[fragment].coding_method != MODE_COPY )
            {
                /* do not perform left edge filter for left columns frags */
                if (x > 0) {
                    s->dsp.vp3_h_loop_filter(
                        plane_data + s->all_fragments[fragment].first_pixel,
                        stride, bounding_values);
                }

                /* do not perform top edge filter for top row fragments */
                if (y > 0) {
                    s->dsp.vp3_v_loop_filter(
                        plane_data + s->all_fragments[fragment].first_pixel,
                        stride, bounding_values);
                }

                /* do not perform right edge filter for right column
                 * fragments or if right fragment neighbor is also coded
                 * in this frame (it will be filtered in next iteration) */
                if ((x < width - 1) &&
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
                    s->dsp.vp3_h_loop_filter(
                        plane_data + s->all_fragments[fragment + 1].first_pixel,
                        stride, bounding_values);
                }

                /* do not perform bottom edge filter for bottom row
                 * fragments or if bottom fragment neighbor is also coded
                 * in this frame (it will be filtered in the next row) */
                if ((y < height - 1) &&
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
                    s->dsp.vp3_v_loop_filter(
                        plane_data + s->all_fragments[fragment + width].first_pixel,
                        stride, bounding_values);
                }
            }

            fragment++;
        }
    }
}

/**
 * called when all pixels up to row y are complete

    if (slice >= s->macroblock_height)
        return;

    for (plane = 0; plane < 3; plane++) {
        uint8_t *output_plane = s->current_frame.data    [plane];
        uint8_t *  last_plane = s->   last_frame.data    [plane];
        uint8_t *golden_plane = s-> golden_frame.data    [plane];
        int stride            = s->current_frame.linesize[plane];
        int plane_width       = s->width  >> !!plane;
        int plane_height      = s->height >> !!plane;
        int y =        slice *  FRAGMENT_PIXELS << !plane ;
        int slice_height = y + (FRAGMENT_PIXELS << !plane);

        /* for each fragment row in the slice (both of them)... */
        for (; y < slice_height; y += 8) {

            /* for each fragment in a row... */
            for (x = 0; x < plane_width; x += 8, i++) {

                if ((i < 0) || (i >= s->fragment_count)) {
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:render_slice(): bad fragment number (%d)\n", i);
                    return;
                }

                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
                        motion_source= golden_plane;
                    else
                        motion_source= last_plane;

                    motion_source += s->all_fragments[i].first_pixel;
                    motion_halfpel_index = 0;

                    /* sort out the motion vector if this fragment is coded
                     * using a motion vector method */
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&

                           put_no_rnd_pixels_l2 which would look more like the
                           VP3 source but this would be slower as
                           put_no_rnd_pixels_tab is better optimzed */
                        if(motion_halfpel_index != 3){
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
                                output_plane + s->all_fragments[i].first_pixel,
                                motion_source, stride, 8);
                        }else{
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
                            s->dsp.put_no_rnd_pixels_l2[1](
                                output_plane + s->all_fragments[i].first_pixel,
                                motion_source - d,
                                motion_source + stride + 1 + d,
                                stride, 8);
                        }
                        dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];

                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
                            block[0] += 128<<3;
                        s->dsp.idct_put(
                            output_plane + s->all_fragments[i].first_pixel,
                            stride,
                            block);
                    } else {
                        s->dsp.idct_add(
                            output_plane + s->all_fragments[i].first_pixel,
                            stride,
                            block);
                    }
                } else {

                    /* copy directly from the previous frame */
                    s->dsp.put_pixels_tab[1][0](
                        output_plane + s->all_fragments[i].first_pixel,
                        last_plane + s->all_fragments[i].first_pixel,
                        stride, 8);

                }
            }
            // Filter the previous block row. We can't filter the current row yet

      */

    // now that we've filtered the last rows, they're safe to display
    if (slice)
        vp3_draw_horiz_band(s, 16*slice);
}

/*
 * This function computes the first pixel addresses for each fragment.
 * This function needs to be invoked after the first frame is allocated
 * so that it has access to the plane strides.
 */
static void vp3_calculate_pixel_addresses(Vp3DecodeContext *s)
{
#define Y_INITIAL(chroma_shift)  s->flipped_image ? 1  : s->fragment_height >> chroma_shift
#define Y_FINISHED(chroma_shift) s->flipped_image ? y <= s->fragment_height >> chroma_shift : y > 0

    int i, x, y;
    const int y_inc = s->flipped_image ? 1 : -1;

    /* figure out the first pixel addresses for each of the fragments */
    /* Y plane */
    i = 0;
    for (y = Y_INITIAL(0); Y_FINISHED(0); y += y_inc) {
        for (x = 0; x < s->fragment_width; x++) {
            s->all_fragments[i++].first_pixel =
                s->golden_frame.linesize[0] * y * FRAGMENT_PIXELS -
                    s->golden_frame.linesize[0] +
                    x * FRAGMENT_PIXELS;
        }
    }

    /* U plane */
    i = s->fragment_start[1];
    for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) {
        for (x = 0; x < s->fragment_width / 2; x++) {
            s->all_fragments[i++].first_pixel =
                s->golden_frame.linesize[1] * y * FRAGMENT_PIXELS -
                    s->golden_frame.linesize[1] +
                    x * FRAGMENT_PIXELS;
        }
    }

    /* V plane */
    i = s->fragment_start[2];
    for (y = Y_INITIAL(1); Y_FINISHED(1); y += y_inc) {
        for (x = 0; x < s->fragment_width / 2; x++) {
            s->all_fragments[i++].first_pixel =
                s->golden_frame.linesize[2] * y * FRAGMENT_PIXELS -
                    s->golden_frame.linesize[2] +
                    x * FRAGMENT_PIXELS;
        }
    }
}

/*
 * This is the ffmpeg/libavcodec API init function.
 */

    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
    s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
    s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
    s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
    s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int));
    s->pixel_addresses_initialized = 0;
    if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts ||
        !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) {
        vp3_decode_end(avctx);
        return -1;
    }

            return -1;
        }

        /* golden frame is also the current frame */
        s->current_frame= s->golden_frame;

        /* time to figure out pixel addresses? */
        if (!s->pixel_addresses_initialized)
        {
            vp3_calculate_pixel_addresses(s);
            s->pixel_addresses_initialized = 1;
        }
    } else {
        /* allocate a new current frame */
        s->current_frame.reference = 3;
        if (!s->pixel_addresses_initialized) {
            av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
            return -1;
        }
        if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");

        return -1;
    }
    if (unpack_dct_coeffs(s, &gb)){
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
        return -1;
    }

    s->last_slice_end = 0;
    for (i = 0; i < s->macroblock_height; i++)
        render_slice(s, i);

11133:cd2956d08cc1

} Coeff;

//FIXME split things out into their own arrays
typedef struct Vp3Fragment {
    Coeff *next_coeff;
    /* this is the macroblock that the fragment belongs to */
    uint16_t macroblock;
    uint8_t coding_method;
    int8_t motion_x;
    int8_t motion_y;

    Vp3Fragment *all_fragments;
    uint8_t *coeff_counts;
    Coeff *coeffs;
    Coeff *next_coeff;
    int fragment_start[3];
    int data_offset[3];

    ScanTable scantable;

    /* tables */
    uint16_t coded_dc_scale_factor[64];

    /* this is a list of indexes into the all_fragments array indicating
     * which of the fragments are coded */
    int *coded_fragment_list;
    int coded_fragment_list_index;

    /* track which fragments have already been decoded; called 'fast'
     * because this data structure avoids having to iterate through every
     * fragment in coded_fragment_list; once a fragment has been fully
     * decoded, it is removed from this list */

    int height          = s->fragment_height >> !!plane;
    int fragment        = s->fragment_start        [plane] + ystart * width;
    int stride          = s->current_frame.linesize[plane];
    uint8_t *plane_data = s->current_frame.data    [plane];
    if (!s->flipped_image) stride = -stride;
    plane_data += s->data_offset[plane] + 8*ystart*stride;

    for (y = ystart; y < yend; y++) {

        for (x = 0; x < width; x++) {
            /* This code basically just deblocks on the edges of coded blocks.

            if( s->all_fragments[fragment].coding_method != MODE_COPY )
            {
                /* do not perform left edge filter for left columns frags */
                if (x > 0) {
                    s->dsp.vp3_h_loop_filter(
                        plane_data + 8*x,
                        stride, bounding_values);
                }

                /* do not perform top edge filter for top row fragments */
                if (y > 0) {
                    s->dsp.vp3_v_loop_filter(
                        plane_data + 8*x,
                        stride, bounding_values);
                }

                /* do not perform right edge filter for right column
                 * fragments or if right fragment neighbor is also coded
                 * in this frame (it will be filtered in next iteration) */
                if ((x < width - 1) &&
                    (s->all_fragments[fragment + 1].coding_method == MODE_COPY)) {
                    s->dsp.vp3_h_loop_filter(
                        plane_data + 8*x + 8,
                        stride, bounding_values);
                }

                /* do not perform bottom edge filter for bottom row
                 * fragments or if bottom fragment neighbor is also coded
                 * in this frame (it will be filtered in the next row) */
                if ((y < height - 1) &&
                    (s->all_fragments[fragment + width].coding_method == MODE_COPY)) {
                    s->dsp.vp3_v_loop_filter(
                        plane_data + 8*x + 8*stride,
                        stride, bounding_values);
                }
            }

            fragment++;
        }
        plane_data += 8*stride;
    }
}

/**
 * called when all pixels up to row y are complete

    if (slice >= s->macroblock_height)
        return;

    for (plane = 0; plane < 3; plane++) {
        uint8_t *output_plane = s->current_frame.data    [plane] + s->data_offset[plane];
        uint8_t *  last_plane = s->   last_frame.data    [plane] + s->data_offset[plane];
        uint8_t *golden_plane = s-> golden_frame.data    [plane] + s->data_offset[plane];
        int stride            = s->current_frame.linesize[plane];
        int plane_width       = s->width  >> !!plane;
        int plane_height      = s->height >> !!plane;
        int y =        slice *  FRAGMENT_PIXELS << !plane ;
        int slice_height = y + (FRAGMENT_PIXELS << !plane);

        /* for each fragment row in the slice (both of them)... */
        for (; y < slice_height; y += 8) {

            /* for each fragment in a row... */
            for (x = 0; x < plane_width; x += 8, i++) {
                int first_pixel = y*stride + x;

                if ((i < 0) || (i >= s->fragment_count)) {
                    av_log(s->avctx, AV_LOG_ERROR, "  vp3:render_slice(): bad fragment number (%d)\n", i);
                    return;
                }

                        (s->all_fragments[i].coding_method == MODE_GOLDEN_MV))
                        motion_source= golden_plane;
                    else
                        motion_source= last_plane;

                    motion_source += first_pixel;
                    motion_halfpel_index = 0;

                    /* sort out the motion vector if this fragment is coded
                     * using a motion vector method */
                    if ((s->all_fragments[i].coding_method > MODE_INTRA) &&

                           put_no_rnd_pixels_l2 which would look more like the
                           VP3 source but this would be slower as
                           put_no_rnd_pixels_tab is better optimzed */
                        if(motion_halfpel_index != 3){
                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
                                output_plane + first_pixel,
                                motion_source, stride, 8);
                        }else{
                            int d= (motion_x ^ motion_y)>>31; // d is 0 if motion_x and _y have the same sign, else -1
                            s->dsp.put_no_rnd_pixels_l2[1](
                                output_plane + first_pixel,
                                motion_source - d,
                                motion_source + stride + 1 + d,
                                stride, 8);
                        }
                        dequantizer = s->qmat[s->all_fragments[i].qpi][1][plane];

                    if (s->all_fragments[i].coding_method == MODE_INTRA) {
                        if(s->avctx->idct_algo!=FF_IDCT_VP3)
                            block[0] += 128<<3;
                        s->dsp.idct_put(
                            output_plane + first_pixel,
                            stride,
                            block);
                    } else {
                        s->dsp.idct_add(
                            output_plane + first_pixel,
                            stride,
                            block);
                    }
                } else {

                    /* copy directly from the previous frame */
                    s->dsp.put_pixels_tab[1][0](
                        output_plane + first_pixel,
                        last_plane + first_pixel,
                        stride, 8);

                }
            }
            // Filter the previous block row. We can't filter the current row yet

      */

    // now that we've filtered the last rows, they're safe to display
    if (slice)
        vp3_draw_horiz_band(s, 16*slice);
}

/*
 * This is the ffmpeg/libavcodec API init function.
 */

    s->all_fragments = av_malloc(s->fragment_count * sizeof(Vp3Fragment));
    s->coeff_counts = av_malloc(s->fragment_count * sizeof(*s->coeff_counts));
    s->coeffs = av_malloc(s->fragment_count * sizeof(Coeff) * 65);
    s->coded_fragment_list = av_malloc(s->fragment_count * sizeof(int));
    s->fast_fragment_list = av_malloc(s->fragment_count * sizeof(int));
    if (!s->superblock_coding || !s->all_fragments || !s->coeff_counts ||
        !s->coeffs || !s->coded_fragment_list || !s->fast_fragment_list) {
        vp3_decode_end(avctx);
        return -1;
    }

            return -1;
        }

        /* golden frame is also the current frame */
        s->current_frame= s->golden_frame;
    } else {
        /* allocate a new current frame */
        s->current_frame.reference = 3;
        if (!s->golden_frame.data[0]) {
            av_log(s->avctx, AV_LOG_ERROR, "vp3: first frame not a keyframe\n");
            return -1;
        }
        if(avctx->get_buffer(avctx, &s->current_frame) < 0) {
            av_log(s->avctx, AV_LOG_ERROR, "vp3: get_buffer() failed\n");

        return -1;
    }
    if (unpack_dct_coeffs(s, &gb)){
        av_log(s->avctx, AV_LOG_ERROR, "error in unpack_dct_coeffs\n");
        return -1;
    }

    for (i = 0; i < 3; i++) {
        if (s->flipped_image)
            s->data_offset[i] = 0;
        else
            s->data_offset[i] = ((s->height>>!!i)-1) * s->current_frame.linesize[i];
    }

    s->last_slice_end = 0;
    for (i = 0; i < s->macroblock_height; i++)
        render_slice(s, i);